On 8/11/2022 6:02 PM, Miriam Taza wrote:

External Email - Use Caution

Hi again,

I have a few questions regarding this.

indeed it does appear that there is a bias field in all of these scans, but this is not specific to those that fail sclimbic.

why do you think it is consistently only the left side that fails? My question here is if whether it is mainly the left due to the way sclimbic does its magic or if this might be due to the actual data (i,e smaller left volumes might have an impact).

I have no idea. That's the problem with the machine learning stuff, it just gives you an answer without a reason.

2. I would really appreciate if someone could explain what --percentile 99.9 is doing a bit more as my results differ a lot depending if this flag is used or not.

Im asking because Im not sure what is the best input to use for sclimbic anymore....

The --percentile 99.9 tells it to sort all the voxels by intensity. One of those voxels will rank the 99.9% in intensity. All voxels that are brighter than this intensity are then set to this intensity ("clipping"). This represents more closely how ScLimbic was trained. I will make --percentile 99.9 the default on the next release.

I decided to look at differences(correlation) of volume (.stats) between sclimbic or sclimbic with --percentile 99.9 ran on T1s or on recon all results (I wanted to make confirm the results using recon all subj vs t1s with the percentile flag are similar given that recon all does correct for bias field). below are the correlation and means. this was on subsample (n=180) and I only looked at one ROI.

left BF right BF

sclimbic x sclimbic99 0.48 0.62

sclimbic x recon99 0.30 0.55

sclimbic x recon 0.30 0.55

recon99 x recon 0.99 0.99

recon99 x sclimbic99 0.91 0.95

recon x sclimbic99 0.92 0.95

avg left BF avg BF mean

sclimbic 198.77 253.96

sclimbic99 297.99 328.93

recon 312.12 340.44

recon99 315.80 343.92

(recon = sclimbic ran on recon all outputs; recon99 = sclimbic --percentile 99 ran on recon outputs; sclimbic=ran sclimbic on t1s; sclimbic99=with flag ran on t1s)

3.so the most important question is what would you judge to provide the "most accurate" result to be used in subsequent group analysis. recon99, recon, sclimbic, or sclimbic99.

This confirms my suspicions above (ie, the training). The "recon" method will most closely track the training (and the results from the paper).

Thank you very much!

Miriam

From: freesurfer-bounces@nmr.mgh.harvard.edu <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of Miriam Taza <miriam.taza@mail.mcgill.ca>
Sent: Wednesday, August 10, 2022 1:26 PM
To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>
Subject: Re: [Freesurfer] sclimbic sometimes does not detect left basal forebrain
External Email - Use Caution

but this(bright at back and nose) also seems to be the case with participants that sclimbic did work on..

From: freesurfer-bounces@nmr.mgh.harvard.edu <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of Douglas N. Greve <dgreve@mgh.harvard.edu>
Sent: Tuesday, August 9, 2022 10:34 PM
To: freesurfer@nmr.mgh.harvard.edu <freesurfer@nmr.mgh.harvard.edu>
Subject: Re: [Freesurfer] sclimbic sometimes does not detect left basal forebrain
There was something a little strange about the cases that were failing -- there were very bright values at the back of the head and around the nose. This probably messes up the normalization in the unet. I ran it with --percentile 99.9 (this eliminates the extreme voxels) and the results looked ok after that.

On 7/28/2022 8:58 PM, Miriam Taza wrote:
Nothing strange about them.

They’re T1w scans 1x1x1mm3

From: freesurfer-bounces@nmr.mgh.harvard.edu <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of Douglas N. Greve <dgreve@mgh.harvard.edu>
Sent: Thursday, July 28, 2022 1:25 PM
To: freesurfer@nmr.mgh.harvard.edu <freesurfer@nmr.mgh.harvard.edu>
Subject: Re: [Freesurfer] sclimbic sometimes does not detect left basal forebrain
Have you looked at the input volume to see if there is anything strange about them? What kind of scans are you passing to it? What is the resolution?

On 7/25/2022 9:03 PM, Miriam Taza wrote:
External Email - Use Caution

Hello,

I noticed 20 out of ~240 subjects left basal forebrain volumes were not picked up after running ScLimbic. These are healthy adults and their scans look good. Also, I noticed often when this occurs NAcc is also 0 or other regions.

I am concerned if there is a systematic issue with left volumes ending up smaller than they should relative to right.

Thanks,
Miriam
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

	left BF	right BF
sclimbic x sclimbic99	0.48	0.62
sclimbic x recon99	0.30	0.55
sclimbic x recon	0.30	0.55
recon99 x recon	0.99	0.99
recon99 x sclimbic99	0.91	0.95
recon x sclimbic99	0.92	0.95

	avg left BF	avg BF mean
sclimbic	198.77	253.96
sclimbic99	297.99	328.93
recon	312.12	340.44
recon99	315.80	343.92